Simulation methods finance_2

Introduction to Monte Carlo in Finance
Giovanni Della Lunga
WORKSHOP IN QUANTITATIVE FINANCE
Bologna - May 12-13, 2016
Giovanni Della Lunga (WORKSHOP IN QUANTITATIVE FINANCE)Introduction to Monte Carlo in Finance Bologna - May 12-13, 2016 1 / 79

Valuation of American Option
Outline
1 Valuation of American Option
2 The Hull and White Model
3 MCS for CVA Estimation
Deﬁnitions
CVA of a Plain Vanilla Swap: the Analytical Model
CVA of a Plain Vanilla Swap: the Simulation Approach

Valuation of American Option by Simulation
As we have seen Monte Carlo simulation is a flexible and powerful
numerical method to value financial derivatives of any kind.
However being a forward evolving technique, it is per se not suited to
address the valuation of American or Bermudan options which are
valued in general by backwards induction.
Longstaff and Schwartz provide a numerically efficient method to
resolve this problem by what they call Least-Squares Monte Carlo.
The problem with Monte Carlo is that the decision to exercise an
American option or not is dependent on the continuation value.

Consider a simulation with M + 1 points in time and I paths.
Given a simulated index level St,i , t ∈ {0, ..., T}, i ∈ {1, ..., I}, what is
the continuation value Ct,i (St,i ), i.e. the expected payoﬀ of not
exercising the option?
The approach of Longstaﬀ-Schwartz approximates continuation values
for American options in the backwards steps by an ordinary
least-squares regression.
Equipped with such approximations, the option is exercised if the
approximate continuation value is lower than the value of immediate
exercise. Otherwise it is not exercised.

In order to explain the metodology, let’s start from a simpler problem.
Consider a bermudan option which is similar to an american option,
except that it can be early exercised once only on a speciﬁc set of
dates.
In the next ﬁgure, we can represent the schedule of a put bermudan
option with strike K and maturity in 6 years. Each year you can
choose whether to exercise or not ...

Let’s consider a simpler example: a put option which can be exercised
early only once ...

Can we price this product by means of a Monte Carlo? Yes we can!
Let’s see how.
Let’s implement a MC which actually simulates, besides the evolution
of the market, what an investor holding this option would do (clearly
an investor who lives in the risk neutral world). In the following
example we will assume the following data, S(T) =, K =, r =, σ =,
t1 = 1y, T = 2y.
We simulate that 1y has passed, computing the new value of the
asset and the new value of the money market account
S(t1 = 1y) = S(t0)e(r−1
2
σ2)(t1−t0)+σ
√
t1−t0N(0,1)
B(t1 = 1y) = B(t0)er(t1−t0)

At this point the investor could exercise. How does he know if it is
convenient?
In case of exercise he knows exactly the payoﬀ he’s getting.
In case he continues, he knows that it is the same of having a
European Put Option.
So, in mathematical terms we have the following payoﬀ in t1
max [K − S(t1), P(t1, T; S(t1), K)]
where P(t1, T; S(t1), K) is the price of a Put which we compute
analytically! In the jargon of american products, P is called the
continuation value, i.e. the value of holding the option instead of
early exercising it.

So the premium of the option is the average of this discounted payoﬀ
calculated in each iteration of the Monte Carlo procedure.
1
N
i
max [K − Si (t1), P(t1, T; Si (t1), K)]
Some considerations are in order.
We could have priced this product because we have an analytical
pricing formula for the put. What if we didn’t have it?

Brute force solution: for each realization of S(t1) we run another
Monte Carlo to price the put.
This method (called Nested Monte Carlo) is very time consuming.
For this very simple case it’s time of execution grows as N2, which
becomes prohibitive when you deal with more than one exercise date!
Let’s search for a ﬁner solution analyzing the relationship between the
continuation value (in this very simple example) and the simulated
realization of S at step t1.
let’s plot the discounted payoﬀ at maturity, Pi , versus Si (t1) ...

As you can see, the analytical price of the put is a curve which kinds
of interpolate the cloud of Monte Carlo points.
This suggest us that the price at time t1 can be computed by means
of an average on all discounted payoff (i.e. the barycentre of the
cloud made of discounted payoff)
So maybe... the future value of an option can be seen as the problem
of finding the curve that best fits the cloud of discounted payoff (up
to date of interest)!!!
In the next slide, for example, there is a curve found by means of a
linear regression on a polynomial of 5th order...

We now have an empirical pricing formula for the put to be used in
my MCS
P(t1, T, S(t1), K) = c0 + c1S(t1) + c2S(t1)2
+ c3S(t1)3
+ c4S(t1)4
+ c5S(t1)5
The formula is obviously fast, the cost of the algorithm being the best
ﬁt.
Please note that we could have used any form for the curve (not only
a plynomial).
This method has the advantage that it can be solved as a linear
regression, which is fast.

The Longstaﬀ-Schwartz Algorithm
The major insight of Longstaﬀ-Schwartz is to estimate the
continuation value Ct,i by ordinary least-squares regression, therefore
the name ”Least Square Monte Carlo” for their algorithm;
They propose to regress the I continuation values Yt,i against the I
simulated index levels St,i .
Given D basis functions b with b1, . . . , bD : RD → R for the
regression, the continuation value Ct,i is according to their approach
approximated by:

Given D basis functions b with b1, . . . , bD : RD → R for the
regression, the continuation value Ct,i is according to their approach
approximated by:
ˆCt,i =
D
d=1
αd,tbd (St,i ) (1)
The optimal regression parameters αd,t are the result of the
minimization
min
α1,t ,...,αD,t
1
I
I
i=1
Yt,i −
D
d=1
αd,tbd (St,i )
2

Simulate I index level paths with M + 1 points in time leading to
index level values St,i , t ∈ {0, ..., T}, i ∈ {1, ..., I};
For t = T the option value is VT,i = hT (ST,i ) by arbitrage
Start iterating backwards t = T − ∆t, . . . , ∆t:
regress the Tt,i against the St,i , i ∈ {1, . . . , I}, given D basis function b
approximate Ct,i by ˆCt,i according to (1) given the optimal parameters
αd,t from (2)
set
Vt,i =
ht(St,i ) if ht(St,i ) > ˆCt,i exercise takes place
Yt,i if ht(St,i ) ≤ ˆCt,i no exercise takes place
repeat iteration steps until t = ∆t;
for t = 0 calculate the LSM estimator
ˆV LSM
0 = e−r∆t 1
I
I
i=1
V∆t,i

Notebook
GitHub : polyhedron-gdl;
Notebook : mcs american;

The Hull and White Model
Outline
Deﬁnitions

Hull-White Model
Described by the SDE for the short rate
dr = (θ(t) − ar)dt + σdw (1)
See Brigo-Mercurio ...
Our version simpliﬁed: a and σ constant;
AKA Extended Vasicek (Note: r(t) is Gaussian);
θ determined uniquely by term structure;

Hull-White Model: Solving for r(t)
d(eat
r) = eat
dr + aeat
rdt = θ(t)eat
+ eat
σdw
integrating both sides we obtain
eat
r(t) = r(0) +
t
0
θ(s)eas
ds + σ
t
0
eas
dw(s)
simplify
r(t) = r(0)e−at
+
t
0
θ(s)e−a(t−s)
ds + σ
t
0
e−a(t−s)
dw(s)

Hull-White Model: Solving for P(t,T)
P(t, T) = V (t, r(t)) where V solves the PDE
Vt + (θ(t) − ar)Vr +
1
2
σ2
Vrr − rV = 0
Final-time condition V (T, r) = 1 for all r at t = T;
Ansatz:
V = A(t, T)e−B(t,T)r(t)
A and B must satisfy:
At − θ(t)AB +
1
2
σ2
(AB)2
= 0, and Bt − aB + 1 = 0
Final-time conditions
A(T, T) = 1 and B(T, T) = 0

Hull-White Model: Solving for P(t,T)
B independent of θ so
B(t, T) =
1
a
1 − e−a(T−t)
(2)
Solving for A requires integration of θ
A(t, T) = exp

−
T
t
θ(s)B(s, T)ds −
σ2
2a2
(B(t, T) − T + t) −
σ2
4a
B(t, T)2



Hull-White Model: Determining θ
Determining θ from the term structure at time 0;
Goal: demonstrate the relation
θ(t) =
∂f
f ∂T
(0, t) + af (0, t) +
σ2
2a
(1 − e−2at
) (3)
Recall
f (t, T) = −∂ log P(t, T)/∂T

We have
− log P(0, T) =
T
0
θ(s)B(s, T)ds +
σ2
2a2
[B(0, T) − T] +
σ2
4a
B(0, T)2
+ B(0, T)r0
Diﬀerentiating and using that B(T, T) = 1 and ∂T B − 1 = −aB we
get
f (0, T) =
T
0
θ(s)∂T B(s, T)ds−
σ2
2a2
B(0, T)+
σ2
2a2
B(0, T)∂T B(0, T)+∂T B(0, T)r0

Diﬀerentiating again, get:
∂T f (0, T) = θ(T) +
T
0
θ(s)∂TT B(s, T)ds
−
σ2
2a2
∂T B(0, T)
+
σ2
2a2
[(∂T B(0, T))2
+ B(0, T)∂TT B(0, T)]
+ ∂TT B(0, T)r0
(4)

Combine these equations, and use a∂T B + ∂TT B = 0;
Get:
af (0, T) + ∂T f (0, T) = θ(T) −
σ2
2a
(aB + ∂T B) +
σ2
2a
[aB∂T B + (∂T B)2
+ B∂TT B]
Substitute formula for B and simplify to get
af (0, T) + ∂T f (0, T) = θ(T) −
σ2
2a
(1 − e−2aT
)
QED

Additive Factor Gaussian Model
The model is given by dynamics (Brigo-Mercurio p. 143):
r(t) = x(t) + φ(t)
where
dx(t) = −ax(t)dt + σdWt x(0) = 0
and φ is a deterministic shift which is added in order to ﬁt exactly the
initial zero coupon curve

So the short rate r(t) is distributed normally with mean and variance
given by (Brigo-Mercurio p.144 equations 4.6 with η = 0)
E(rt|rs) = x(s)e−a(t−s)
+ φ(t)
Var(rt|rs) =
σ2
2a
1 − e−2a(t−s)
where φ(T) = f M(0, T) + σ2
2a 1 − e−aT 2
and f M(0, T) is the
market instantaneous forward rate at time t as seen at time 0.

Model discount factors are calculated as in Brigo-Mercurio (section
4.2):
P(t, T) =
PM(0, T)
PM(0, t)
exp (A(t, T))
A(t, T) =
1
2
[V (t, T) − V (0, T) + V (0, t)] −
1 − e−a(T−t)
a
x(t)
where
V (t, T) =
σ2
a2
T − t +
2
a
e−a(T−t)
−
1
2a
e−2a(T−t)
−
3
2a

MCS for CVA Estimation
Outline
Deﬁnitions

MCS for CVA Estimation Definitions
Counterparty Risk
In general a given company, say a financial institution A, will have
portfolios with many other counterparties, varying among sovereign
entities, corporates, hedge funds, insurance companies;
Counterparty credit exposure is the amount a company, say A, could
potentially lose in the event of one of its counterparties defaulting.
As we’ll see in more details, it can be computed by simulating in
different scenarios and at different times in the future, the price of the
transactions with the given counterparty, and then by using some
chosen statistic to characterise the price distributions that have been
generated.

CVA Context
Counterparty risk can be considered broadly from two different points
of view;
The first point of view is risk management, leading to capital
requirements revision, trading limits discussions and so on;
The second point of view is valuation or pricing, leading to amounts
called Credit Valuation Adjustment (CVA) and extensions thereof,
including netting, collateral, re-hypothecation, close-out specification,
wrong way risk and funding cost;

Example: CVA of plain vanilla Interest-Rate Swap
Consider counterparties A and B who enter into an interest-rate swap
where A receives every six months the 6-month Libor rate on a
notional of 100 million euro, while paying to B a ﬁxed amount equal
to the par 10-year swap rate on the same notional observed at
inception.
This is a typical swap contract with value zero at inception.
As time passes and market conditions change, the value of the swap
changes accordingly. Thus, if the swap rate decreases (resp.
increases), the transaction will be out of the money (resp. in the
money) as seen from A’s point of view.

Example: CVA of plain vanilla Interest-Rate Swap
Therefore, if B were to default at a point in the life of the trade when
swap rates had increased, then A would need to replace in the
market—at higher cost than the fixed amount being paid to B—the
floating cashflows promised and not delivered by B.
To compute the credit exposure for the swap, we would need to
estimate the values the swap could take in different market scenarios
at points in the future.
For practical reason it can be useful to characterise the distribution of
values with some quantity which can be useful for various risk
controlling;
For example we could compute the so called Expected Positive
Exposure (more in the following) as
EPEt = E[V +
t ]

Exposure Proﬁle: EPE
Figure shows the EPE of the swap price distribution, over its entire
life, as seen from party A’s point of view.

Exposure Proﬁle: Value Density

Exposure Proﬁle
Figure shows the full price
distribution over time.
The fair value for the swap, and
this must therefore have value
(and hence exposure level)
identically equal to zero at
inception.
Similarly, towards the end of the
transaction, when all payments
but one due under the swap
have been paid, the exposure
remaining is that from only a
single coupon exchange.

Exposure Profile
This explains what happens at the
right end of the profile.
At intermediate times, the shape of
the profile is the result of opposing
effects.
On the one hand, as the interest rates
underlying the swap diffuse, there is
more variability in the realised Libor
rates, potentially leading to higher
exposure.
On the other hand, as time evolves
there are fewer payments remaining
under the swap, and this mitigates the
effect of diffusing rates.

Exposure Profiles
So there are two main effects that determine the credit exposure over
time for a single transaction or for a portfolio of transactions with the
same counterparty: diffusion and amortization;
as time passes, the ”diffusion effect” tends to increase the exposure,
since there is greater variability and, hence, greater potential for
market price factors to move significantly away fro current levels;
the ”amortization effect”, in contrast, tends to decrease the exposure
over time, because it reduces the remaining cash flows that are
exposed to default;

Exposure Profiles
For single cash flow products, such as FX Forwards, the potential
exposure peaks at the maturity of the transaction, because it is driven
purely by ”diffusion effect”;
on the other hand, for products with multiple cash flows, such as
interest-rate swaps, the potential exposure usually peaks at one-third
to one-half of the way into the life of the transaction;

Credit Value Adjustment
Credit Value Adjustment (CVA) is by deﬁnition the diﬀerence
between the risk-free portfolio value and the true portfolio value that
takes into account the possibility of a counterparty’s default;
in other words CVA is the market value of counterparty credit risk;

Assuming independence between exposure and counterparty’s credit
quality greatly simpliﬁes the analysis.
under this assumption we can write
CVA = (1 − R)
T
0
EE (t)dPD(0, t) (5)
where EE (t) is the risk-neutral discounted expected exposure (EE)
given by
EE (t) = EQ B0
Bt
E(t) (6)

In general calculating discounted EE requires simulations;
Exposure is simulated at a ﬁxed set of simulation dates, therefore the
integral in (5) has to be approximated by the sum:
CVA = (1 − R)
N
i=1
EE (tk)PD(tk−1, tk) (7)
Since expectation in (6) is risk-neutral, scenario models for all price
factors should be arbitrage free;
This is achieved by appropriate calibration of drifts and volatilities
speciﬁed in the price-factor evolution model;

MCS for CVA Estimation CVA of a Plain Vanilla Swap: the Analytical Model
The General Unilateral Counterparty Risk Pricing Formula
At valuation time t, and provided the counterparty has not defaulted
before t, i.e. on {τ > t}, the price of our payoﬀ with maturity T
under counterparty risk is
Et[¯Π(t, T)] = Et[Π(t, T)] − Et[LGD It≤τ≤T D(t, τ)(NPV (τ))+
]
positive counterparty-risk adjustment
= Et[Π(t, T)] − UCVA(t, T)
(8)
with
UCVA(t, T) = Et[LGD It≤τ≤T D(t, τ)(NPV (τ))+
]
= Et[LGD It≤τ≤T D(t, τ)EAD]
(9)
Where LGD = 1 − REC is the loss given default, and the recovery
fraction REC is assumed to be deterministic.

It is clear that the value of a defaultable claim is the sum of the value
of the corresponding default-free claim minus a positive adjustment.
The positive adjustment to be subtracted is called (Unilateral) Credit
Valuation Adjustment (CVA), and it is given by a call option (with
zero strike) on the residual NPV at default, giving nonzero
contribution only in scenario where τ ≤ T.

Counterparty risk thus adds and optionality level to the original payoff.
This renders the counterparty risk payoff model dependent even when
the original payoff is model independent.
This implies, for example, that while the valuation of swaps without
counterparty risk is model independent, requiring no dynamical model
for the term structure (no volatility and correlations in particular), the
valuation of swaps under counterpaty risk will require and interest
rate model.

Now we explore the well known result that the component of the IRS
price due to counterparty risk is the sum of swaption prices with
different maturities, each weighted with the probability of defaulting
around that maturity.
Let us suppose that we are a default free counterparty ”B” entering a
payer swap with a defaultable counterparty ”C”, exchanging fixed for
floating payments at times Ta+1, . . . , Tb.
Denote by βi the year fraction between Ti−1 and Ti , and by P(t, Ti )
the default-free zero coupon bond price at time t for maturity Ti . We
take a unit notional on the swap.
The contract requires us to pay a fixed rate K and to receive the
floating rate L resetting one period earlier until the default time τ of
”B” or until final maturity T if τ > T.
The fair forward-swap rate K at a given time t in a default-free
market is the one which renders the swap zero-valued in t.

In the risk-free case the discounted payoﬀ for a payer IRS is
b
i=a+1
D(t, Ti )βi (L(Ti−1, Ti ) − K) (10)
and the forward swap rate rendering the contract fair is
K = S(t; Ta, Tb) = Sa,b(t) =
P(t, Ta) − P(t, Tb)
b
i=a+1
P(t, Ti )βi

Of course if we consider the possibility that ”C” may default, the
correct spread to be paid in the ﬁxed leg is lower as we are willing to
be rewarded for bearing this default risk.
In particular we have
UCVA(t, Tb) = LGD Et[Iτ≤Tb
D(t, τ)(NPV (τ))+
]
= LGD
Tb
Ta
PS (t; s, Tb, K, S(t; s, Tb), σs,Tb
) dsQ{τ ≤ s}
(11)
being PS (t; s, Tb, K, S(t; s, Tb), σs,Tb
) the price in t of a swaption
with maturity s, strike K underlying forward swap rate S(t; s, Tb),
volatility σs,Tb
and underlying swap with ﬁnal maturity Tb.

The proof is the following: given independence between τ and the
interest rates, and given that the residual NPV is a forward start IRS
starting at the default time, the option on the residual NPV is a sum
of swaptions with maturities ranging the possible values of the default
time, each weighted (thanks to the independence assumption) by the
probabilities of defaulting around each time value.
We can simplify formulas allow the default to happen only at points
Ti of the ﬁxed leg payment grid.
In this way the expected loss part is simpliﬁed.

Indeed in the case of postponed (default occour to the ﬁrst Ti
following τ) payoﬀ we obtain:
UCVA(t, Tb) = LGD Et [Iτ≤Tb
D(t, τ)(NPV (τ))+
]
= LGD
b−1
i=a+1
PS (t; s, Tb, K, S(t; s, Tb), σs,Tb ) (Q{τ ≥ Ti } − Q{τ > Ti })
(12)

Notebook
Notebook : n09 cva swap;

MCS for CVA Estimation CVA of a Plain Vanilla Swap: the Simulation Approach
Monte Carlo Framework for CVA
The industry practice to compute exposure is to use a simple Monte Carlo
framework implemented in three steps:
1 scenario generation,
2 pricing
3 aggregation.

The ﬁrst step involves generating scenarios of the underlying risk
factors at future points in time.
Simple products can then be priced on each scenario and each time
step, therefore generating empirical price distributions.
From the price distribution at each time it is then possible to extract
convenient statistical quantities.
Exposure of portfolios can be computed by consistently pricing
diﬀerent products on the same underlying scenarios and aggregating
the results taking into account possible netting and collateral
agreement with the counterparty.

If taken literally, this approach works only for relatively simple
products which can be priced analytically, or which can be
approximated in analytical form, and which do not need complex
calibrations depending on market scenarios.
More exotic products requiring relatively complex pricing, cannot be
treated in this way.

1 Simulate yield curve at future dates
2 Calculate your derivatives portfolio NPV (net present value) at each
time point for each scenario
3 Calculate CVA as sum of Expected Exposure multiplied by probability
of default at this interval
CVA = (1 − R) DF(t)EE(t)dQt
where R is the Recovery Rate (normally set to 40%) EE(t) is the
expected exposure at time t and dQt the survival probability density,
DF(t) is the discount factor at time t.

In this simple example we will use a modiﬁed version of Hull White
model to generate future yield curves. In practice many banks use
some yield curve evolution models based on this model.
For each point of time we will generate whole yield curve based on
short rate. Then we will price our interest rate swap on each of these
curves;
To approximate CVA we will use BASEL III formula for regulatory
capital charge approximating default probability [or survival
probability ] as exp(−ST /(1 − R)) so we get
CVA = (1 − R)
i
EE(Ti ) + EE(Ti−1
2
e−S(Ti−1)/(1−R)
− e−S(Ti )/(1−R)
where EE is the discounted Expected Exposure of portfolio.

Notebook
Notebook :
n10 mcs cva swap;

Implementation Challenges
The Monte Carlo framework we have shown in the previous section,
seems to give a good implementation recipe.
For a given portfolio of transactions we could identify the underlying
risk factors and simulate forward (or spot) prices, taking into account
correlations if required, use functions already implemented to price
each product, and then derive statistical quantities.
As we have mentioned already, this could be the approach followed by
a ﬁnancial institution to assess the counterparty credit risk of its OTC
derivatives portfolios.

In the implementation phase, however, there can be issues which need
to be addressed.
The generation of correlated scenarios is not trivial, as there can be
thousands of diﬀerent risk factors driving the dynamics of products in
the portfolio.
Consider for example an equity portfolio, where each underlying stock
needs, at least in principle, a speciﬁc simulation.

Not all products can be computed in analytical form.
Most exotics are priced on grids using PDEs or using Monte Carlo
approaches.
In these cases the exposure computation would require a Monte Carlo
simulation for scenarios and a Monte Carlo simulation, or a PDE
computation, for each scenario and time step to price the instrument.
This becomes quickly unfeasible from a computational point of view.
In addition, depending on the model used for pricing, calibration could
also become problematic, as it has to be performed at each scenario.

American Monte Carlo
The points highlighted in the previous section clearly show that the
classical Monte Carlo scheme has intrinsic limitations and that we
need an alternative approach.
There are possibilities to circumvent in a systematic way some of the
problems related to valuation and architecture.
The basic idea is to approach the counterparty exposure problem as a
pricing problem, and thus to use pricing algorithms, which generate
not just the value of a trade at inception, but rather a price
distribution at predetermined time steps.
One possibility is to use the so called American Monte Carlo
algorithm, which we will refer to as, simply, the AMC algorithm.
The main feature of this algorithm is that, instead of building a price
moving forward in time, it starts from maturity, where the value of
the transaction is known, and goes backwards, till the inception.

AMC CVA Simulation
Let’s illustrate with a Bermuda Swaption which can be exercised at 2 dates, T1
and T2.

AMC CVA Simulation
Forward Phase: we diffuse risk factors and store in memory contract cash flows
and payoff variables as the will be used in the Backward Phase.

AMC CVA Simulation
Backward Phase: we discount payoﬀ contract and add cash ﬂows up to T2.

AMC CVA Simulation
Backward Phase: At T2 we obtain a cloud of points...

AMC CVA Simulation
... regression is performed to obtain product fair value at T2 as a function of
regression basis.

AMC CVA Simulation
We discount to T1 obtaining a cloud of points (adding present cash ﬂow from
forward phase).

AMC CVA Simulation
Regression is performed at T1 as a function of regression basis.

AMC CVA Simulation
Forward Phase: we then move forward again and invoke regression functions with
the new set of MC paths ...

AMC CVA Simulation
We can compute distribution of MTMs, exposure, EPE, PFE, etc...

AMC CVA Simulation
Recap
Important Remark: In Phase 3, MC diﬀusion can be diﬀerent from
Phase 1.

Credits
Damiano Brigo, Fabio Mercurio ”Interest Rate Models — Theory and
Practice” Springer Finance (2006)
Damiano Brigo, Massimo Morini, Andrea Pallavicini ”Counterparty
Credit Risk, Collateral and Funding” Wiley Finance (2013)
Umberto Cherubini, Elisa Luciano, Walter Vecchiato ”Copula
Methods in Finance” Wiley Finance (2004)
Tommaso Gabbriellini ”American Options with Monte Carlo”
Presentation on Slideshare
Yves Hilpisch ”Derivatives Analytics with Python” Wiley Finance
(2015)
Don L. McLeish ”Monte Carlo Simulation and Finance” Wiley
Finance (2005)

Simulation methods finance_2

More Related Content

What's hot

Viewers also liked

Similar to Simulation methods finance_2

More from Giovanni Della Lunga

Recently uploaded

Simulation methods finance_2